We present a new framework for image segmentation with statistical shape model enhanced level sets represented as a linear combination of non-Euclidean radial basis functions (RBFs). The shape prior for the level set is represented as a probabilistic map created from the training data and registered with the target image. The new framework has the following advantages: 1) the explicit RBF representation of the level set allows the level set evolution to be represented as ordinary differential equations and reinitialization is no longer required. 2) The non-Euclidean distance RBFs makes it possible to incorporate image information into the basis functions, which results in more accurate and topologically more flexible solutions. Experimental results are presented to demonstrate the advantages of the method, as well as critical analysis of level sets versus the combination of both methods.